This work is devoted to the modification of neural network method for classifying the helicopters turboshaft engines ratings at flight modes using neural network technologies, which, through the use of a new hybrid network of ART-1 and BAM, made it possible to improve the quality of recognition of operating modes to almost 100 %. The hybrid network ART-1 and BAM training process was modified, which made it possible to adapt the network without adding a new class and train it to recognize existing classes when the incoming data only slightly differs from those recorded in long-term memory. This makes it possible to associate non-identical data with one identifier vector, which makes it possible, when using the classifier in helicopters turboshaft engines automatic control system, to correctly respond to the presented data.